一句话：提出用大语言模型为 HPC 库自动生成测试驱动，以突破差分测试中人工用例覆盖不足与现有自动化技术失效的困境。
高性能计算（HPC）依赖 BLAS、FFTW 等硬件相关库，其正确性通常通过差分测试来验证。然而，人工编写测试用例覆盖有限；符号执行与模糊测试等自动化技术又因 HPC 库中存在隐式参数约束、且难以生成合法的大规模输入，导致效率低下、自动化程度不足。该论文针对这一痛点，提出以大语言模型驱动测试驱动程序的生成，试图在差分测试框架下自动产出可用的测试驱动，从而提升测试覆盖与自动化水平。摘要仅阐述了问题背景与现有方法的局限，未展开具体生成机制与实验评估；其核心贡献可概括为将 LLM 引入 HPC 库测试驱动的构造环节。
High-performance computing (HPC) relies on hardware-specific libraries like BLAS and FFTW, whose correctness is often validated via differential testing. However, manual test cases limit coverage, and automated techniques such as symbolic execution and fuzzing struggle with HPC libraries due to implicit parameter constraints and the difficulty of generating valid, large-scale inputs, leading to inefficiency and poor automation.
@inproceedings{paperbot11987,
title = {Large Language Model powered Test Driver Generation for High-performance Computing Library},
author = {Ziran He and Guofeng Zhang and Meixi Liu and Zhenbang Chen},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3786582.3786815}
}